Internal models that banks use to assess the creditworthiness of their borrowers, as a rule, give estimates of the probability of default that cover the entire business cycle. For the purposes of applying IFRS 9, however, estimates of the probability of default for a specific moment, as well as the inclusion of different macroeconomic scenarios are required. Such estimates are based primarily on the calculation of the effects of the business cycle, and therefore involve the existence of a provable link between macroeconomic indicators and realized default rates. In this paper we analyze whether this relationship exists using the data of banks operating in Serbia. We use several different approaches to determine this link: linear regression, autoregressive process, error correction model, static and dynamic panel-data models, as well as two Bayesian approaches. On the whole sample, the error correction model shows the best performance and gives the factors of acceptable economic intuition. When data are divided by the type of product, we obtain somewhat less reliable results, which is partly due to the dominant influence of the SME segment in the total default rates. As the most robust predictors of default rates we identify the lagged differences in these rates, the reference rate of the National Bank of Serbia and the growth rate of gross domestic product.